TDMA frame, Time slot and speech data transfer confusion.

Hi guys, I read Goller's documentation "About the GSM-Dm-Channels", but I'm a bit confused about something. I understand that one TDMA frame has a period of 4.615 ms and consists of 8 Time slots, each of a period of 576.9 us. When a MS registers to the BTS, it gets one Time-slot, but how is speech transfered? Normally the Fs (samplefrequency) is 8 KHz, this means every 128 us one byte sample. If I have one timeslot of let's say 577 us , than there is a gap of 7 other timeslots, which makes a 4 ms gap. Also the speech data is compressed. How is that damn speech data multiplexed with other MSs and finally received on the other side as 8 bit samplevalue at a rate of 128 us (8 KHz). It gets fuzzy to me, cause I wrote a microcontroller project which, among other things, samples audio and than transfered to a GSM module. This is done at 8 KHz samplerate, so how is that done when I have only one TimeSlot of 577 us every 4 ms period? I can't find a good link about how several MSs on one ARFCN can send and receive speech data. Thank you.